Multimodal and Data-Driven Approaches in Acupuncture Research: Methods, Applications, and Challenges

This comprehensive review examines the digital revolution in acupuncture research, marking a fundamental transition from traditional hypothesis-driven methods to data-driven approaches. The authors present a structured framework that categorizes acupuncture data into three main components: intervention data (stimulation parameters, point selection), response data (clinical outcomes, multimodal biomarkers), and contextual data (traditional Chinese medicine diagnosis, patient characteristics). Causal analysis emerged as an essential tool for overcoming the limitations of conventional statistical methods. Mixed-effects models demonstrated value in multicenter studies, controlling between-center variability and repeated measures.

Bayesian methods proved superior in small samples with complex outcomes, while causal inference techniques such as propensity score matching and difference-in-differences enabled more rigorous analyses of observational and real-world data. The target trial emulation framework represents an important methodological advance for studies with electronic health records. Artificial intelligence transformed the ability to analyze heterogeneous and multidimensional data. Supervised machine learning algorithms, including support vector machines (SVMs), LASSO, and tree-based methods, demonstrated accuracy of 77-88% in predicting treatment response.

Deep learning significantly expanded the processing of unstructured data: convolutional neural networks (CNNs) analyzed neuroimaging and tongue/pulse diagnostic images, while recurrent networks (RNNs) and transformers modeled temporal signals such as EEG and needle manipulation. Variational autoencoders revealed low-dimensional neural subspaces associated with acupuncture, suggesting that the technique reorganizes the dynamic landscape of cortical activity. Text analysis, including natural language processing and large language models (LLMs), enabled extraction of structured knowledge from classical texts and contemporary records. Bibliometric tools mapped the evolution of the field, identifying emerging areas such as neurological rehabilitation.

Integrative analyses combined multi-omics data, multimodal neuroimaging, and physiological signal fusion, providing a systemic understanding of mechanisms of action. Studies revealed modulation of metabolic pathways, neural connectivity, and central-peripheral coupling, elucidating molecular and neurophysiological bases of therapeutic effects. Practical applications span four main domains: efficacy assessment using robust causal models, efficacy prediction through ML algorithms, mechanistic investigation via neuroimaging analysis and omics data, and clinical decision support through objective diagnosis and AI-assisted prescription. Studies demonstrated that routine clinical characteristics can predict responders with accuracy above 77%, while neuroimaging biomarkers reached 85% accuracy.

Mechanistic analyses revealed that different manipulation techniques produce distinct patterns of cortical activation and reorganization of neural networks. Clinical support systems integrated objective imaging diagnosis, automated point localization, and manipulation quantification, facilitating a transition beyond dependence on subjective experience. Despite the promising advances, significant limitations persist. The lack of data standardization represents a fundamental obstacle, with substantial heterogeneity in collection protocols and representation of contextual data.

Many models remain exploratory, with limited external validation and restricted interpretability. Methodological complexity requires multidisciplinary expertise, limiting practical implementation. Issues of data quality, model generalization, and clinical translation have not yet been adequately resolved. The implications for the future of acupuncture research are transformative.

These methods facilitate a shift from population-level average effects to individualized strategies, allowing analysis of treatment heterogeneity and underlying mechanisms. The integration of multimodal data with sophisticated algorithms promises precision medicine in acupuncture, where therapeutic decisions are informed by individual profiles of biomarkers, clinical characteristics, and contextual factors. The development of standardized pipelines linking data acquisition, analytical validation, and clinical application represents a crucial priority for translating these methodological advances into tangible benefits for patients.